ai_trend: groupBy gpu type

kernelboard/api/leaderboard.py

@@ -303,14 +303,28 @@ def parse_model_from_filename(file_name: str) -> str:
 
 def group_by_model(items: List[dict]) -> dict:
     """
-    Group time series items by model name for easier charting.
+    Group time series items by gpu_type first, then by model name.
+    Skips items where model or gpu_type is unknown.
+    Returns: { "H100": { "claude-opus-4.5": [...], "gpt-5": [...] } }
     """
     series = {}
     for item in items:
-        model = item.get("model", "unknown")
-        if model not in series:
-            series[model] = []
-        series[model].append({
+        gpu_type = item.get("gpu_type")
+        model = item.get("model")
+
+        # Skip if model or gpu_type is unknown or missing
+        if not gpu_type or gpu_type == "unknown":
+            continue
+        if not model or model == "unknown":
+            continue
+
+        if gpu_type not in series:
+            series[gpu_type] = {}
+
+        if model not in series[gpu_type]:
+            series[gpu_type][model] = []
+
+        series[gpu_type][model].append({
             "submission_time": item["submission_time"],
             "score": item["score"],
             "submission_id": item["submission_id"],